ScamLensCrypto Scam Field Guide
Deepfakes

Deepfake Video-Call Scams: How to Check Who You're Talking To

A video-call screen with a face mask peeled back at one corner and another face underneath
Both the face and the voice on a video can be synthesized. To verify, switch to a contact detail you saved yourself.

At the end of January 2024, Hong Kong police received a report: someone had received an email from the "Chief Financial Officer of the UK head office", been invited into a group video conference, and afterwards, as instructed, moved about HK$200 million into five local bank accounts. Police investigation found that the conference had been generated in advance from publicly available video clips and voices downloaded online (see the Secretary for Security's written reply to the Legislative Council of June 2024). A face on video no longer works as proof of identity. The step that still does: hang up, and call back on a contact detail you saved yourself.

Three points first:
  • Ask the other person to wave a hand, press their nose or turn their head: if the picture breaks, it is fake. If it doesn't, that tells you nothing about whether a real person is there.
  • The check has to start from you. Hang up and call back on a number from your own contacts or through a channel inside the official app, preferably on a different app or by phone.
  • If you have already sent money, keep the original files of any video, voice message or call recording from the other side. Don't forward or convert them.

Where crypto users run into deepfakes

These are the three with a public source behind them.

Synthetic videos of celebrities and "analysts" endorsing an investment scheme

In a video, a public figure or a "chief analyst" recommends a platform, a coin or a "can't-lose" strategy. In a public service announcement dated July 20, 2026, the FBI's Internet Crime Complaint Center (IC3) wrote that scammers use AI-generated videos to create believable depictions of public figures to bolster their fraud schemes. What matters is what you are asked to do after watching: join a group, download a particular app, move money into an account the other side gives you.

A group video meeting with a fake executive

This is the case described at the top. According to the Secretary for Security's written reply, the fraudster first sent a phishing email posing as the Chief Financial Officer of the victim's UK head office, inviting him to a group video conference about confidential transactions. The victim authorized transfers to five local bank accounts as instructed and lost about HK$200 million. Police believe the conference was pre-recorded, so there was no actual interaction: after giving the instructions, the fraudster ended the meeting under a pretext and continued issuing payment instructions through instant messaging. The same reply describes a second case from May 2024, in which a staff member of a multinational company received a WhatsApp message from the "Chief Financial Officer of the UK head office", then joined a video conference and was told to transfer nearly HK$4 million. Neither case involved crypto, and the method works on anyone who has authority to move money.

The "law enforcement officer" who turns up after you have been scammed

You have already been scammed once, and then someone claiming to be from a law enforcement agency contacts you and says they can help get the money back. This is the scheme the IC3 announcement describes: scammers impersonating FBI personnel who handle IC3 complaints, and building spoofed websites with a structure and content similar to the official IC3 site in order to gather personal information. Their targets include previous victims. The announcement also states that IC3 will never directly communicate with individuals by phone, email, social media, apps, online chat or public forums, and will never ask for payment to recover lost funds.

The FBI IC3 public service announcement of July 2026 warning that scammers are impersonating the IC3 and FBI personnel, using AI-generated video and spoofed websites, and targeting previous victims
The FBI IC3 announcement of July 20, 2026 (number I-072026-PSA): AI-generated video, spoofed websites, previous victims targeted. Screenshot taken August 2026.

In all three, the face does the same job: it gets you to skip the check you would otherwise have made.

If the person in the video is talking about "AI quant trading" or "follow the bot for steady profits", you can test the promised return itself with the compound return calculator.

Can asking someone to wave or turn their head expose a deepfake?

One tip is: on a video call, ask the other person to pass a hand across their face, press their nose or turn their head quickly, and watch whether the picture smears or the edges flicker. The tip is usable, but only as a one-way filter.

A visible flaw means fake. No visible flaw does not mean real

If you see blurred edges, garbled ears or hair, lips out of sync with the words, or the picture tearing as a hand crosses the face, treat it as fake. The reverse doesn't hold: seeing no flaw does not show that a real person is there. How well the fake was made, and how closely you happened to be looking, can both let a flaw slip past.

The case at the top shows something else as well. That conference was recorded in advance and had no interaction at all, and the money still went out. Someone being hurried through a "confidential transaction" is not paying attention to the picture. So the method below does not depend on your eyesight. The question it asks is "did I complete the check?", not "can I spot it?"

How to verify that the person on video is who they say they are

When the person on video starts asking for money, for a verification code, or for you to download something, hang up, then call back on a completely separate contact route to check.

This works because what can be synthesized is the content: face, voice, tone, background. What can't be faked is a route in your hands that the other side doesn't control. Once you are the one who starts the contact, their setup can't reach that route.

  • Take the number from your own contacts. Don't redial from the call log. Caller ID can be spoofed, and an account on a messaging app can be a newly registered imitation. Use the entry you saved earlier.
  • For anything involving a platform, go through the official channel inside its app or website. Don't use a link the other person sent, or a support number from search results. To check whether a number or account they gave you is official, use the platform's own verification page. The ones for OKX and Binance are listed in the table under where to download the OKX and Binance apps.
  • Change the medium. Don't "confirm again" inside the same app. If they can impersonate someone on a messaging app, they can intercept your checking message there too. They started a video call, so you ring back by phone. The account is on Telegram, so you call the person's real mobile number.
  • Ask about something small that only the real person knows and that has never been online. For example, "what was the name of the place we ate at last time?" Don't ask anything that can be found on their social media. This is a supplement. Calling back is the main step.
  • For payment instructions inside a company, make two-person approval a fixed rule. Any payment instruction given verbally in a video meeting needs a second confirmation in person or over another route, so that it is never left to judgment in the moment.

When you can't see a flaw, what do you go on?

On what the other person asks you to do next. A face can be swapped. The actions the second half of a scam needs from you cannot:

What you see or hearWhat it meansWhat to do
An instruction during the video to transfer money, pay on someone's behalf or lend coinsA payment instruction is the strongest signal there is, whoever appears to be giving itHang up and call back on another route to confirm
You are told to download an "exclusive app" or an "internal version"What it can do once installed is up to the other sideEnd the call. If you do need a platform's app, open its official website and follow the link from there to the app store
You are asked to share your screen so they can "show you how"What they want to see is your verification codes and account pagesRefuse, whatever reason is given
You are told to move money into a "safe account" or an "exclusive account"Money moved into an account someone else names is no longer in your handsHang up
Someone claims to be law enforcement and says they can recover the money you lostIC3 states that it will never directly contact individuals or ask for payment to recover lost fundsDo nothing on that call. Ring the authority that took your report yourself
You are hurried throughout: "do it now", "the meeting is waiting"You are being given no time to checkThe more they push, the more reason to stop. Hang up first

If any row of the table applies, hang up and call back another way. There is no need to study the face any further.

One kind of footage needs a separate mention: the other person shows you that they "just passed the platform's face verification". That is a clip played for your benefit. The platform has sent you no confirmation, and the whole "verification successful" sequence could be a screen recording or a synthetic clip, so it is not evidence that a real person is there. In the other direction, don't send clear, front-facing video of yourself to people you don't know, and don't perform requested movements on camera to "help with verification".

Isn't AI-generated video supposed to be labeled? Does no label mean it's real?

No. China's Measures for Labeling AI-Generated Synthetic Content, issued by the Cyberspace Administration and three other departments and in force from September 1, 2025, contain two requirements. Service providers must add an explicit label to generated content (text or a corner mark that users can see) and an implicit label in the file's metadata (content attributes, provider information and so on). For an individual trying to tell real from fake, its usefulness has limits:

  • The measures regulate those who provide generation services. Someone setting out to defraud will not label their own video, and an existing label can be removed.
  • So a label works in one direction only. If you see one, you know the content is synthetic. If you don't, nothing follows.
  • Metadata gets lost as content circulates. After forwarding, compression, screen recording or re-editing, the implicit label may be gone, so "I checked the metadata and found no label" is not evidence either.

When a transfer is about to be made, the step to use is still the same: hang up and call back to check.

You already sent money to the person on video: the extra evidence to keep

The order for limiting the damage and reporting is in Scammed in Crypto? What to Do and Where to Report It. A case like this has one more thing to keep:

Keep the original video file. Don't convert it, and don't keep only screenshots

If you have video or voice messages the other side sent you, or a screen recording of the call, keep the original files exactly as they are. Don't forward them to someone through a messaging app (they may be re-compressed on the way), don't keep only screenshots, and don't convert the format to save space. The encoding information and metadata in the original may have forensic value and can be lost in a single round of compression. Save a copy of the original locally or to cloud storage before you do anything else.

If you aren't sure whether what is in front of you is a scam, run it through the scam self-check.

Public sources cited

  • FBI Internet Crime Complaint Center, FBI Warns of Scammers Impersonating the IC3 (number I-072026-PSA, July 20, 2026): impersonation of FBI personnel handling IC3 complaints, use of AI-generated video and spoofed websites, previous victims among the targets.
  • Government of the Hong Kong Special Administrative Region, press release, LCQ9: Combating frauds involving deepfake (written reply by the Secretary for Security in the Legislative Council, June 26, 2024): the case of about HK$200 million reported at the end of January 2024 and the case of nearly HK$4 million in May 2024.
  • Cyberspace Administration of China and three other departments, Measures for Labeling AI-Generated Synthetic Content (in Chinese): in force from September 1, 2025. The detailed requirements for explicit and implicit labels are in the text of the measures.

Related pages